Individual Preferences for COVID-19 Vaccination in China

Anli Leng1,Elizabeth Maitland2, Stephen Nicholas3, Rugang Liu4 ,Jian Wang5,6*

1 Anli Leng School of Political Science and Public Administration, Institute of Governance,
Shandong University,72 Binhai Rd, Qingdao, Shandong 266237,China; email:[email protected]

2 Elizabeth Maitland School of Management, University of Liverpool, Chatham Building, Chatham

Street, Liverpool L697ZH, England; email:[email protected]

3 Stephen Nicholas Australian National Institute of Management and Commerce,1 Central

Avenue Australian Technology Park, Eveleigh Sydney NSW 2015, Australia; Newcastle Business
School, University of Newcastle, University Drive, Newcastle, NSW 2308, Australia;
email:[email protected]

4 Rugang Liu School of Health Policy & Management, Nanjing Medical University, No. 101
Longmian Avenue, Jiangning District, Nanjing, 211166, PRC; email: [email protected]

5 Jian Wang Dong Fureng Institute of Economic and Social Development, Wuhan University, 54
Dongsi Lishi Hutong, Beijing, 100010, China; email:[email protected]; Tel:+8613864157135
6 Center for Health Economics and Management at School of Economics and Management, Wuhan
University, Wuhan University, Wuhan, Hubei Province, P.R.China. 430072
* corresponding author

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 Research in context
Evidence before this study
Large vaccination coverage is seen as the effective choice to stop the pandemic of COVID-19. With the
hope that a COVID-19 vaccine will be developed in the near future, it is vital to understand
individuals’ vaccination preferences and vaccination decision-making in order to predict vaccination
coverage, to assess the factors influencing vaccination decision-making, and to improve vaccination
coverage.There is little research on individuals’ COVID-19 vaccination decision-making. Previous
vaccination studies on the HBV, seasonal influenza and HPV identified various vaccine-specific factors
that influence an individual’s vaccination preferences, such as the vaccine’s side effects, effectiveness,
duration of protection, cost, number of doses, transmission route, location of vaccination sites,
willingness to pay. There is no agreement on the importance of these vaccine-specific aspects on the
willingness to vaccinate. Also, there is no consensus on whether vaccine-led herd immunity leads to
higher vaccine acceptance through altruistic motives or hinders individual vaccinations through the
free-rider problem. No studies have identified whether inter-individual vaccination preference
heterogeneity may impact COVID-19 vaccine uptake depending on individuals’ socio-demographic
characteristics and attitudes to vaccine.

Added value of this study

To our knowledge, this is the first study to estimate individual preferences for a COVID-19 vaccine for
a nationally representative population in China. Data was selected from a D-efficient discrete choice
experiment which was conducted across 6 Chinese provinces selected by the stratified random
sampling method. Although all seven attributes (vaccine effectiveness, side-effects, proportion of
acquaintances vaccinated, vaccine protection, number of doses, access to vaccine and vaccination sites)
were proved to significantly influence respondents’ vaccination decision, vaccine effectiveness, side-
effects and proportion of acquaintances vaccinated were the most important.We also found a higher
probability of vaccinating when the vaccine was more effective; risks of serious side effects were small;
vaccinations were free and voluntary; the fewer the number of doses; the longer the protection duration;
and the higher the proportion of acquaintances vaccinated. Higher local vaccine coverage created
altruistic herd incentives to vaccinate rather than free-rider problems. The predicted vaccination uptake
of the optimal vaccination scenario in our study was 84.77%. Preference heterogeneity was substantial.
Individuals who were older, had a lower education level, lower income, higher trust in the vaccine and
higher perceived risk of infection, displaing a higher probability to vaccinate.

Implications of all the available evidence

To maximize COVID-19 vaccine uptake, health authorities should promote vaccine effectiveness; pro-
actively communicate the absence or presence of vaccine side effects; and ensure rapid and wide media
communication about local vaccine coverage. Preference heterogeneity among individuals should lead
health authorities to address the diversity of expectations about COVID-19 vaccinations by providing
people-oriented immunization services.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 Individual Preferences for COVID-19 Vaccination in China

Summary

Background: Vaccinations are an effective choice to stop disease outbreaks, including COVID-19.
There is little research on individuals’ COVID-19 vaccination decision-making. In this study, we aimed
to determine individual preferences for COVID-19 vaccinations in China, and to assess the factors
influencing vaccination decision-making to facilitate vaccination coverage.

Methods: A D-efficient discrete choice experiment was conducted across 6 Chinese provinces selected
by the stratified random sampling method.Vaccine choice sets were constructed using seven attributes:
vaccine effectiveness, side-effects, accessibility, number of doses, vaccination sites, duration of
vaccine protection, and proportion of acquaintances vaccinated. Conditional logit and latent class
models were used to identify preferences.

Findings:Although all seven attributes were proved to significantly influence respondents’ vaccination
decision, vaccine effectiveness, side-effects and proportion of acquaintances vaccinated were the most
important.We also found a higher probability of vaccinating when the vaccine was more effective; risks
of serious side effects were small; vaccinations were free and voluntary; the fewer the number of doses;
the longer the protection duration; and the higher the proportion of acquaintances vaccinated.Higher
local vaccine coverage created altruistic herd incentives to vaccinate rather than free-rider
problems.The predicted vaccination uptake of the optimal vaccination scenario in our study was
84.77%.Preference heterogeneity was substantial.Individuals who were older, had a lower education
level, lower income, higher trust in the vaccine and higher perceived risk of infection, displaing a
higher probability to vaccinate.

Interpretation:Preference heterogeneity among individuals should lead health authorities to address the
diversity of expectations about COVID-19 vaccinations.To maximize COVID-19 vaccine uptake,
health authorities should promote vaccine effectiveness; pro-actively communicate the absence or
presence of vaccine side effects; and ensure rapid and wide media communication about local vaccine
coverage.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 Introduction

By the end of July, 2020, COVID-19 had infected more than 16 million people across
216 countries or territories, with more than 65,000 deaths worldwide1. During the pandemic, there have
been no specific antiviral drugs to treat COVID-19 effectively, but existing drugs used to treat other
viral diseases, such as dexamethasone, have improved the recovery of high-risk COVID-19 patients. A
vaccine is seen as the effective choice to stop the pandemic, with more than 100 COVID-19 vaccines in
development worldwide2, including seven that have now been approved for human testing through
clinical trials. With the hope that a COVID-19 vaccine will be developed in the near future, it is vital to
understand individuals’ vaccination preferences and vaccination decision-making in order to predict
vaccination coverage, to assess the factors influencing vaccination decision-making, and to improve
COVID-19 vaccination coverage. There is little research on individuals’ COVID-19 vaccination
decision-making. To addresses this lacuna, we test an individual choice COVID-19 vaccination model
across 6 provinces in China.

Previous vaccination studies identified various vaccine-specific factors that influence an individual’s
vaccination preferences, such as the vaccine’s side effects, effectiveness, duration of protection, cost,
number of doses, transmission route, location of vaccination sites, willingness to pay and the burden of
the disease3-7. There is no agreement on the importance of these vaccine-specific aspects on the
willingness to vaccinate. Also, there is no consensus on whether vaccine-led herd immunity leads to
higher vaccine acceptance through altruistic motives6,8 or hinders individual vaccinations through the
free-rider problem9. Further, inter-individual vaccination preference heterogeneity may impact vaccine
uptake depending on individuals’ socio-demographic characteristics, such as age6, education10 and
income11,12. Also, trust in the vaccine and social trust may be important factors in individual
heterogeneity, because they influence people’s decision to vaccinate, especially given that many
vaccines have a reputation for poor quality, leading to a drop in vaccine coverage13,14.

Widely employed in studies of the HBV5, seasonal influenza4 and human papillomavirus vaccines7, we
used the discrete choice experiment (DCE) method to reveal and measure preferences for a potential
COVID-19 vaccine. The conditional logit model (CLM) was used to test whether respondents'
predefined characteristics alter the mean preference associated with attribute values. In addition, the
latent class model (LCM) was used to estimate inter-individual preference heterogeneity. Our study is
one of the first attempts to determine the preferences of individuals in China for a COVID-19 vaccine.
While a China-based study, the research will inform other countries on individual COVID-19 decision-
making, facilitating vaccination coverage in line with China and World Health Organization goals.

Method

Identification of Attributes and Levels

Identification of vaccine attributes and their levels is key for ensuring the validity of DCE. We
retrieved relevant vaccine attributes and levels from the literature3-7,10,15. Attributes were then ranked,
categorized and reduced through interviews with a group of eight experts in the field of vaccination, an
eight person focus group discussion and a 30 survey pilot study. We identified seven key attributes and

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 adopted a DCE design with 8 choice sets of two vaccine profiles.

The details of the attributes and levels are displayed in Table 1. Vaccine effectiveness refer to the
proportion of vaccinated persons protected by the vaccine, with three levels (40%, 60% and 85%),
based on the effectiveness of seasonal influenza vaccinations16,17 and high effectiveness of H1N1
vaccination18. Side effects were assigned three levels: 50/100000, 10/100000 and 1/100000. These
levels were chosen to represent vaccines with moderate side effects, such as seasonal influenza
vaccines16,17 and low side effects, such as H1N1 vaccine19. Access to the vaccine was specified as free
and voluntary (reflecting H1N1 vaccination), free and compulsory (reflecting most current universal
childhood vaccinations) and chargeable and voluntary (reflecting some partial or non-reimbursed
seasonal influenza vaccines). Number of doses was described by two levels: one dose and two or more
doses. Vaccination sites were described by three possible levels: village clinic or community health
station, township or community health centre, and county hospital and above. The duration of vaccine
protection was assigned three levels: 6 months, 1 year and two or more years. These duration levels
were retrieved based on the duration of vaccine protection of seasonal influenza vaccinations16,17,
H1N1 vaccination18 and hepatitis B vaccine5. Acquaintances vaccinated was assigned 30%, 60% and
90% of close acquaintances (friends and family) already being vaccinated. This attribute was to
quantify the importance of local coverage as a main driver for vaccination choices6.

Table 1. Attributes and Levels Used in the Discrete Choice Experiment

Attributes Levels Descriptions
40% Protects 40% of vaccinated
Vaccine effectiveness 60% Protects 60% of vaccinated
85% Protects 85% of vaccinated
50/100000 50 out of 100,000 risk of severe side effect
Vaccine related side-
10/100000 10 out of 100,000 risk of severe side effect
effects
1/100000 1 out of 100,000 risk of severe side effect
Free and voluntary Vaccine is free and it is voluntary to get vaccinated.
Free and compulsory Vaccine is free and it is compulsory to get vaccinated
Access to vaccine
Chargeable and You must pay for vaccine yourself, but vaccination is
voluntary voluntary.
one dose One dose
Number of doses
≥2 doses Two or more doses
First level village clinic or community health station
Vaccination sites Second level township or community health centre
Third level county hospital and above
six months Six months of vaccine protection
Duration of vaccine
one year One year of vaccine protection
protection
More than two years More than two years
30% of your family, friends and acquaintances already
30%
vaccinated
Acquaintances vaccinated
60% of your family, friends and acquaintances already
60%
vaccinated

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 90% of your family, friends and acquaintances already
90%
vaccinated

Experimental Design

To guarantee that preference parameters can be estimated with maximal precision, the D-efficient
partial profile design was used. Forty-eight hypothetical two-alternative choice tasks were created. For
each choice task, respondents were asked to choose which vaccine they would prefer. To reduce the
cognitive burden on respondents, these 48 choice tasks were divided randomly into 6 different versions
of the questionnaire. As shown in the examples in Figure 1, each version included 8 choice tasks.

Survey

A three-part questionnaire sought information on respondents’ background characteristics, attitudes to

COVID-19 and the COVID-19 vaccine, and DCE preferences. Background characteristics comprised
sex, age, location, educational attainment, job status, and income. Using a five-point Likert scale,
attitudes to COVID-19 were surveyed by two questions and attitudes to the COVID-19 vaccine were
surveyed by six questions as show in Appendix 1 Supplemental Material. Vaccine and COVID-19
related attitudes were tested as covariates with the DCE estimates to examine preference heterogeneity.

A pilot survey was conducted with 30 respondents. Based on the pilot, the survey was subsequently
revised to improve phraseology and question layout. Also, the pilot allowed the survey to be adjusted
to ease the cognitive load and to increase the quality of the responses, with 48 pairs of scenarios in the
questionnaires divided into 6 questionnaires with 8 pairs of scenarios in each. To promote survey
accuracy, our DCE started with a general description and an illustrative example of a simplified choice
set to familiarize the respondents with the choice tasks ahead. The study was approved by Nanjing
Medical University Ethics Committee (No.2020565) and the face-to-face survey conducted by trained
researchers.

Using a stratified random sampling method, six Chinese provinces were chosen based on high, medium
and low GDP per capita, geographically covering east, central and west regions of China. In each
province, three cities were chosen, also divided into high, medium and low GDP per capita, with 314

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 individuals randomly surveyed in each city.

Sample

The inclusion criteria were respondents aged over 18 years old, without cognitive impairments.
According to the research standard proposed by Orme20, the minimum sample size should be 94, with
de Bekker-Grop et al.21 reporting from a DCE literature survey that only 9% of DCE sample sizes were
greater than 10000. Our study collected 1888 questionnaires, with 1883 (99.7%) of respondents
answering all choices. The six versions of the questionnaires were evenly distributed, with version 1
accounting for 17.21%, version 2, 16.46%, version 3, 17.05%, version 4, 16.09%, version 5, 15.93%
and version 6, 17.26% of all surveys. In total, 30128 observations comprised the database.

Data Analysis

Following previous research22, all the variables describing attribute levels were introduced as dummy
variables. The quality of the various discrete choice models was compared on the basis of the Bayesian
information criteria (BIC) and Akaike information criteria (AIC)23. The preferences were estimated
with two models, the conditional logit model (CLM) and latent class model (LCM). The CLM is
commonly used to analyze preferences, with individual utility estimated by equation 1:
U ijs  1effect(60)ijs   2 effect(85)ijs   3sideeffect(10)ijs   4sideeffect(1)ijs
  5access(compusary ) ijs   6 access(charge ) ijs   7 doseijs   8 site(secondlevel ) ijs
  9 site(thirdlevel ) ijs  10 protection(1yr)ijs  11 protection(2yr)ijs  12 acquaintances(60)ijs
 13acquaintances(90)ijs   ijs
（1）

where U ijs is the utility of respondent i for scenario j in the choice set s (here j = 1, 2; s = 1,..,8),  is a

parameter vector relating attribute values and utility levels and  ijs is error of utility.

CLM assumes errors are independent and identically distributed, which limits the analysis of response
heterogeneity and neglects preference heterogeneity. Not conforming to CLM model assumptions,
LCM addresses these problems by classifying individuals into mutually exclusive groups, which
display heterogeneous preferences, according to differences in share values, characteristics and
behavior. The AIC and BIC criteria were used to determine the optimal number of classes24. The
optimal utility function was:
U ijs|c  1|c effect(60)ijs|c   2|c effect(85)ijs|c   3|c sideeffect(10)ijs|c   4|c sideeffect(1)ijs|c
  5|c access(compusary ) ijs|c   6|c access(charge ) ijs|c   7|c doseijs|c   8|c site(secondlevel ) ijs|c
  9|c site(thirdlevel ) ijs|c  10|c protection(1yr)ijs|c  11|c protection(2yr)ijs|c  12|c acquaintances(60)ijs|c
 13|c acquaintances(90)ijs|c   ijs|c
(2)

where U ijs|c represents the utility of respondent i belonging to class segment c for scenario j in choice

sets.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 In addition to the utility function, the final model allowed for several covariates (age, education, region,
income, trust in vaccines, and risk of infection) to enter into the class assignment model. The class
assignment utility function for the final model was:
U ijs|c   0 c  1c agei   2 c educationi   3c regioni
  4 c incomei   5c trusti   6 c riski (3)

Policy analysis was conducted by calculating the probability of vaccine uptake. Utility scores were
used as probabilities and attribute levels were predefined. Probabilities of vaccine uptake were
calculated on the basis of an indirect utility analysis, using equation 1, whereby individual (n) will
choose vaccine within a choice set of 1 out of j sets (j =1, ., j) of options.
 'x j
e
Pi 
e
 'x j
(4)

Results

Respondent characteristics

Table 3 presents the respondents’ characteristics. Of the 1883 respondents, 50.98% were female and
60.33% live in urban areas, which was consistent with the current national population distribution
making the sample broadly nationally representative. Key characteristics include 61.23% in the 18-44
age group; 59.27% were married; about 70% had a high school (9<years of schooling≤12 years) or
higher education; 71% were employed or farmers; and almost 45% were in the lowest income category.
The sample was evenly distributed across the three regions.

Table 2 Characteristics of the study sample (n=1883)

Characteristics n %
Gender
Male 923 49.02
Female 960 50.98
Age
Age 18-44 1153 61.23
Age 45-59 545 28.94
Age 60- 185 9.82
Marital status
unmarried 688 36.54
married 1116 59.27
widowed/divorced 79 4.20
Residence
urban area 1136 60.33
rural area 747 39.67
Education

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 low education (years of schooling≤9 years) 576 30.59
medium education (9<years of schooling≤12 years) 809 42.96
high education (years of schooling>12 years) 498 26.45
Occupation
farmer 395 20.98
employed 957 50.82
other (including students and retired) 531 28.20
Household yearly income
<4500RMB 847 44.98
4500-9000RMB 579 30.75
>9000RMB 457 24.27
Region
east 634 33.67
central 624 33.14
west 625 33.19

Attitudes to vaccine and COVID-19

In general, trust in the vaccine and vaccination process was ranked as high importance, with 81.9%
agreeing or strongly agreeing that the vaccine was safe and 90% of respondents agreeing or strongly
agreeing that vaccinating was very important. The perceived COVID-19 risk was not high: only 24.8%
agreed or strongly agreed that they and their friends and relatives were at risk of COVID-19 and 14.5%
of respondents agreed or strongly agreed that someone in their community will have COVID-19 (See
Appendix 1 in Supplemental Materials).

Estimation of preferences and their heterogeneity

Table 3 presents the results of the conditional logit model. All attributes were shown to be significant
(p<0.01), except for “free and compulsory” and “second level” (township or community health center)
vaccination sites. Vaccine effectiveness and side-effects were the most important attributes, followed
by duration of vaccine protection, proportion of acquaintances vaccinated and access to the vaccine.
Number of doses and vaccination sites were the least significant attributes. Our result show that there
was a higher probability for vaccination take-up when the vaccine was more effective, had small risk of
serious side-effects, was free and voluntary, required fewer doses, had a longer protection duration, and
the higher the proportion of acquaintances who vaccinated. The positive correlation between vaccine
utility and the proportion of acquaintances vaccinated demonstrated the peer influence and altruistic
motives on herd immunity. The vaccination sites showed a significant direct linear relationship with
vaccine utility, with a low trust in primary health care and high preference for municipal and above
hospitals as vaccination sites.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 Table 3 Conditional logit model of respondent preferences
Attribute ß SEa 95% CI
Vaccine effectiveness (reference=40%)
60% 0.389*** 0.026 0.338, 0.440
85% 0.994*** 0.027 0.941, 1.047
Vaccine related side-effects
(reference=50/100000)
10/100000 0.469*** 0.026 0.418, 0.520
1/100000 0.932*** 0.027 0.879, 0.985
Access to vaccine
(reference=free and voluntary)
free and compulsory 0.025 0.026 -0.025, 0.075
chargeable and voluntary -0.279*** 0.026 -0.330, -0.228
Number of doses (reference=one dose)
≥2 doses -0.057*** 0.018 -0.093, -0.022
Vaccination sites (reference=first level)
second level -0.030 0.027 -0.082, 0.022
third level 0.073*** 0.026 0.023, 0.124
Duration of vaccine protection (reference=six
months)
one year 0.152*** 0.025 0.102, 0.202
more than two years 0.257*** 0.026 0.206, 0.309
Acquaintances vaccinated (reference=30%)
60% 0.155*** 0.026 0.104, 0.206
90% 0.257*** 0.026 0.206, 0.307
Model fit
Observations=30128 Prob>chi2=0.000
Respondents=1883 LR chi2 (13)=3064.46
AIC 17017.97
BIC 17234.11
a SE standard errors; ***statistically significant at 0.01%

To estimate preference heterogeneity, latent class models were estimated. We selected the three-class
model with six sociodemographic covariates based on AIC and BIC comparisons across two-, three- or
four-class models, including sociodemographic covariates that can improved the model fit25, model
simplicity and interpretability of class membership. Table 4 shows clearly the different preference
heterogeneity among the three classes, and the class probabilities indicated that 54.8% of respondents
were assigned to class 1, 23.2% to class 2, and 21.9% to class 3. For class 1, respondents had identical
preferences to the mean preferences of the whole sample, with the coefficients and their significance of
attribute levels similar to those estimated by the CLM. Vaccine effectiveness, proportion of
acquaintances vaccinated, side-effects and duration of vaccine protection were also important attributes

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 in the LCM, although their rankings differ from the CLM. Access to vaccine and vaccination sites were
less important than the other attributes, and the number of doses were no longer significant.

For LCM class 2, respondents had some different preferences from those estimated by the CLM and
LCM class 1. The significant difference was that side-effects was the attribute whose level made utility
vary most, followed by access to the vaccine. All attributes, including number of doses, were
significant, except for the “two years level of duration of vaccine protection”. Respondents in class 3
showed large differences in preferences compared to those estimated by the LCM class 1 and class 2.
Vaccine effectiveness had the highest importance, which was higher than the side-effect attribute, and
“free and compulsory”, number of doses and “two years” levels were significant in LCM class 3.

Estimating the probability of belonging to any class based on four individual socio-demographic
characteristics, risk perception and vaccine trust, we found they were significantly associated with the
membership of preference classes, including age, education, region, average monthly household
income, trust in vaccines and risk of infection. Older age respondents had a higher probability of
preferring COVID-19 vaccinations. For LCM class 1, respondents with lower education levels and
lower income levels had a higher probability of preferring the COVID-19 vaccination compared with
those with higher education level and higher income level. Compared with those in east region,
respondents in central and west region had a higher probability of preferring the COVID-19
vaccination. For LCM class 2, individuals who had a higher trust in vaccines and a higher perceived
risk of COVID-19 infection, displayed a higher probability to prefer the COVID-19 vaccination.

Table 4. Latent class logit model of respondent preferences

Attribute Class 1 Class 2 Class 3
ß SEa ß SE ß SE
Vaccine effectiveness
60% 0.202*** 0.042 0.736*** 0.212 2.366*** 0.231
85% 0.505*** 0.049 1.806*** 0.167 4.435*** 0.338
Vaccine related side-effects
10/100000 0.170*** 0.039 3.134*** 0.105 0.790*** 0.168
1/100000 0.366*** 0.043 6.510*** 0.532 0.935*** 0.204
Access to vaccine
free and compulsory 0.043 0.037 -1.007*** 0.190 0.434** 0.180
chargeable and voluntary 0.000*** 0.039 -2.017*** 0.164 0.145 0.176
Number of doses
≥2 doses -0.037 0.027 -0.902*** 0.108 0.218* 0.115
Vaccination sites
second level -0.072** 0.037 0.493** 0.203 0.114 0.180
third level 0.025 0.036 0.844*** 0.163 0.173 0.163
Duration of vaccine protection
one year 0.187*** 0.037 -0.412*** 0.149 0.102 0.151
more than two years 0.309*** 0.036 0.151 0.215 0.606*** 0.196
Acquaintances vaccinated

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 60% 0.221*** 0.036 0.798*** 0.188 -0.161 0.180
90% 0.405*** 0.037 0.522*** 0.168 -0.006 0.163
Class probability model
age 0.467*** 0.119 0.350** 0.135 - -
education -0.257*** 0.093 0.109 0.105 - -
region 0.183** 0.090 0.083 0.098 - -
average monthly household income -0.375*** 0.086 -0.109 0.094 - -
trust in vaccines -0.156 0.104 -0.558*** 0.125 - -
risk of infection -0.094 0.088 -0.228** 0.103 - -
constant 1.707*** 0.353 0.242 0.485 - -
Class probability
Average 0.548 0.232 0.219
Model fit
Observations=30128
Respondents=1883
AIC 16334.25
BIC 16766.54
a SE standard errors; ***statistically significant at the 1%; ** at 5%; and * at 10% level

Probability of take-up
Vaccine uptake probabilities were calculated on the basis of an indirect utility analysis, using CLM.
Figure 2 presents the vaccine uptake probabilities. As a comparative standard, we defined a base
vaccination program: 60% vaccine effectiveness 60%, 10/100000 risk of severe side-effects, free and
voluntary vaccinations, one dose, village clinic or community health station, protection duration of 6
months and 60% of acquaintances vaccinated. The base case is indicated as zero change in the
probability on the X-axis in Figure 2. When the vaccine effectiveness increased from 40% to 85%, the
vaccination uptake increased 45.68% and when the risk of severe side effects fell from 50/1000000 to
1/1000000, the vaccination uptake increased 43.18%. The vaccination uptake rate increased 12.38%
when the proportion of acquaintances vaccinated increased from 30% to 90%.

Taking the base vaccination program, the predicted vaccination uptake was 45.82%. The predicted
vaccination uptake of the worst vaccination scenario, (vaccine effectiveness of 40%, 50/100000 risk of
severe side-effects, chargeable and voluntary to get vaccinated, two dose, vaccination sites of first level,
protection duration of 6 months and acquaintances vaccinated of 30%) was still 23.53%. The predicted
vaccination uptake of the optimal vaccination scenario (vaccine effectiveness of 85% ,1/100000 risk of
severe side-effects, free and voluntary vaccinations, one dose, third level vaccination sites, protection
duration of 2 years and 90% of acquaintances vaccinated) was 84.77%.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 Discussion

To our knowledge, this is the first study to estimate individual preferences for a COVID-19 vaccine
using a DCE for a nationally representative population in China. Vaccine effectiveness, side-effects,
proportion of acquaintances vaccinated, vaccine protection, number of doses, access to vaccine and
vaccination sites were attributes significantly influencing the preferences for COVID-19 vaccinations.
Older age individuals, those with a lower education level, lower income, higher trust in vaccines and
high perceived risk of infection had a higher probability to vaccinate. The predicted vaccination uptake
of the optimal vaccination scenario in our study was 84.77%.

Vaccine effectiveness, side-effects, and proportion of acquaintances vaccinated were most important
attributes, but their rankings differed between CLM and LCM. Consistent with Verelst et al’s6 flu study
and Sadique et al’s26 study of three hypothetical vaccines, side-effects were the most important factor
in the CLM and class 2 LCM. But, vaccine effectiveness was the most important attribute in the class 1
and class 3 LCM, which is consistent with Guo et al.’s5 study on HBV vaccinations and Wong et al’s7
research on human papillomavirus vaccinations. We found the local vaccine coverage (or proportion of
acquaintances vaccinated) was the second most important attribute in class 1 LCM (54.85% of the
sample), and even more important than side-effects. This finding was different from Verelst et al’s6 flu
study where the local vaccine coverage was less important than side-effects, accessibility and burden of
disease attributes. The more acquaintances who vaccinated, the more individuals preferred to vaccinate
against COVID-19. This finding suggests that herd immunity through vaccination will lead to higher
levels of vaccine acceptance through altruistic motives and attenuate the free-rider problem motives.
Rapid and wide media communication about local vaccine coverage can help to improve the
vaccination coverage.

Costs were still an important factor for vaccination preferences. Compared with free vaccinations, a
vaccination charge reduced the probability of vaccination. However, the effects of voluntary or

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 compulsory vaccination are different among three classes of respondents. Under the premise of free
vaccine policy, the effects of voluntary or compulsory vaccination policy needs further study. We
expected that when a medical institution vaccination site was convenient, vaccinations would be more
likely to be taken-up. However, our results showed the opposite. Vaccination sites showed a significant
direct linear relationship with vaccine utility in the CLM and class 2 LCM, which may be explained by
the low trust in primary health care centers compared to more distant, but higher trust, municipal and
above hospitals. Also, vaccination sites might reflect the quality of the vaccination service, with
individuals in China reporting more confidence in municipal and above hospital services compared to
primary-level medical and health care institutions5. Not surprisingly, respondents preferred fewer doses,
in part because fewer visit to a vaccination site saves time and transportation costs, which was in line
with Shono and Kondo4 flu and Guo et al.5 hepatitis B vaccinations findings. The longer the duration
of protection, the more individuals prefer to vaccinate.

Respondents in our study showed preference heterogeneity for COVID-19 vaccinations. Information of
preference heterogeneity and knowing the differences in personal values towards COVID-19
vaccinations will help policy makers to understand individual preferences for vaccinations, which will
promote increased vaccine coverage. For example, respondents who are older, have a lower education
level and reported lower incomes had a higher probability to vaccinate. This finding suggests that older,
poorer and less educated individuals may overestimate the benefits from vaccinations; younger, higher
educated and high income respondents may underestimate the safety of the vaccine. Therefore, pro-
actively communication on side-effects and effectiveness of vaccine is critical to stimulate vaccine
uptake, especially for younger, higher educated and high income respondents.

We found more than 30% of the respondents who did not want to get vaccinated were worried about
the side-effects. We also found the trust in the COVID-19 vaccine and the perceived risk of COVID-19
increased COVID-19 vaccine take-up. One policy implication is that health authorities need campaigns
to shape the knowledge and attitudes towards COVID-19 and the COVID-19 vaccine to address any
cognitive biases and distrust in the public. Further, the evidence of altruistic take-up of the vaccine
should encourage policy makers to launch widespread media communications about local vaccine
coverage.

The study had some limitations. First, an opt-out option was not included in our DCE. This is
consistent with the answers obtained in our pilot survey and with the design of our experiment. Second,
we did not include willingness to pay (WTP) or an out-of-pocket cost attribute in our DCE. To reduce
the cognitive burden of respondents, we offered only the binary “free” versus “chargeable” level for the
vaccine. As a result, we cannot estimate and compare the WTP for different vaccination scenarios,
which should be addressed in future studies.

In conclusion, this study provides evidence-based individual preference data to policy makers in terms
of quality of vaccinations, vaccination accessibility and trust in the vaccine. It is critical to stimulate
vaccine uptake by emphasizing vaccine effectiveness and pro-actively communicating about side-
effects and local vaccine coverage. A higher local vaccine coverage can create an altruistic incentive to
vaccinate. Compared with free vaccinations, charging for the vaccine reduced for 70% of the
respondents their utility of vaccinating, resulting in a lower probability of vaccinations. Under the

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 premise of a free vaccine, the effects of voluntary or compulsory vaccination policy were different
among the three classes of respondents, which requires further study. Preference heterogeneity among
individuals should lead health authorities to address the diversity of expectations about COVID-19
vaccinations by providing people-oriented immunization services.

Contributors
JW and AL contributed towards the article by making substantial contributions to conception and
design. AL contributed towards the article by collecting data and undertaking the statistical analysis,
interpretation of the data, and writing the manuscript. SN,EM and RL engaged in interpreting the
results and writing the paper. All authors read and approved the final version of the manuscript.

Declaration of interests
We declare no competing interests.

Data sharing
The dataset supporting the conclusions of this article is available from the corresponding author on
reasonable request.

Funding
This work was supported by China Postdoctoral Science Foundation (grant number 2019M662392) and
Qingdao Postdoctoral Foundation.

Acknowledgments
The authors are grateful to research students in Shandong University,Nanjing Medical University,
Inner Mongolia Medical University, Ningxia Medical University,Xinxiang Medical University for their
assistance in collecting data. The authors thank the editor and reviewers for suggestions that have
significantly improved the paper.

References

1.WHO.Coronavirus disease (COVID-19) pandemic-27 July 2020.

https://www.who.int/emergencies/diseases/novel-coronavirus-2019 [accessed July 27, 2020]
2.WHO. DRAFT landscape of COVID-19 candidate vaccines–30 April 2020.
https://www.who.int/whodocuments-detail/draft-landscape-of-COVID-19-candidate-
vaccines[accessed May 5,2020].
3.Feng-Cai Z, Yu-Hua L, Xu-Hua G, Li-Hua Hou,et al.,Safety, tolerability, and immunogenicity of a
recombinant adenovirus type-5 vectored COVID-19 vaccine: a dose-escalation,open-label,non-
randomised,first-in-human trial,Lancet 2020; 395:1845-1854.
4. Shono A, Kondo M. Parents' preferences for seasonal influenza vaccine for their children in
Japan.Vaccine 2014;32:5071-5076.
5.Guo N, Zhang G, Zhu D, et al. The effects of convenience and quality on the demand for
vaccination: Results from a discrete choice experiment.Vaccine 2017;35:2848-2854.
6.Verelst F, Willem L, Kessels R, et al. Individual decisions to vaccinate one's child or oneself: A
discrete choice experiment rejecting free-riding motives.Social ence & Medicine 2018;207:106-116.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 7.Wong C, Man K, Ip P, et al. Mothers' Preferences and Willingness to Pay for Human
Papillomavirus Vaccination for Their Daughters: A Discrete Choice Experiment in Hong Kong.
Value in Health 2018; 21:622-629.
8.Laguzet, Laetitia, Turinici, Gabriel. Individual vaccination as Nash equilibrium in a SIR model with
application to the 2009-2010 influenza A (H1N1) epidemic in France.Bulletin of Mathematical
Biology 2015;77:1955-1984.
9.Vietri JT, Li M, Galvani AP, et al. Vaccinating to help ourselves and others.Medical Decision
Making 2012;32:447.
10.Bekker-Grob EWD, Rose JM, Donkers B, et al. Men’s preferences for prostate cancer screening:
a discrete choice experiment. British Journal of Cancer 2013;108:533-541.
11.Buchanan J,Wordsworth S, Schuh A. Patients' preferences for genomic diagnostic testing in
chronic lymphocytic leukaemia: a discrete choice experiment. Patient 2016;9: 525–536.
12.Christine P, Pélissier A, Béjean S. Preference heterogeneity with respect to whole genome
sequencing. A discrete choice experiment among parents of children with rare genetic diseases.
Social Science & Medicine 2018;214:125-132.
13.Miyachi T, Takita M, Senoo Y, et al. Lower trust in national government links to no history of
vaccination. The Lancet 2020;395:31-32.
14.Liu Z, Yang JZ. In the Wake of Scandals: How Media Use and Social Trust Influence Risk
Perception and Vaccination Intention among Chinese Parents.Health Communication
2020;7716:1-12.
15.Christine, Michaels-Igbokwe, Shannon,et al.Individual Preferences for Child and Adolescent
Vaccine Attributes: A Systematic Review of the Stated Preference Literature. Patient 2017;10:687-
700.
16.CDD Centers for Disease Control Vaccine Effectiveness - How Well Does the Flu Vaccine Work?
https://www.cdc.gov/flu/about/qa/vaccineeffect.htm, 2017. [accessed May 14,2017]
17.Kelly H , Carville K , Grant K , et al.Estimation of Influenza Vaccine Effectiveness from Routine
Surveillance Data. Plos One 2009;4: e5079.
18.Yu HJ.The vaccination protection rate of HIN1 vaccine exceeds 85%. Persistence is still under
observation.
https://www.ixueshu.com/document/756eb25879eaa0d6318947a18e7f9386.html[accessed Nov
05,2009] (in Chinese)
19. Deng HH, More than 150 cases of HINI vaccination have been reported abnormal reaction in
more than 300,000 people nationwide. http://cn.chinagate.cn/infocus/2009-
10/10/content_18678045.htm[accessed Oct 10, 2009](in Chinese)
20.B.Orme, Getting Started with Conjoint Analysis: Strategies for Product Design and Pricing
Research (second edition), Research Publishers LLC, Madison.2010 Jan.
21.De Bekker-Grob EW, Donkers B, Jonker MF, et al. Sample Size Requirements for Discrete-
Choice Experiments in Healthcare: a Practical Guide.Patient-patient Centered Outcomes Research
2015;8:373-384.
22.Daly A, Dekker T, Hess S.Dummy coding vs effects coding for categorical variables:
clarifications and extensions. Journal of Choice Modelling 2016;21 :36–41.
23.Hauber AB, Gonzalez JM, Prior T, et al. Statistical Methods for the Analysis of Discrete Choice
Experiments: A Report of the ISPOR Conjoint Analysis Good Research Practices Task Fore.
Value in Health 2016;19:300-315.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226
 24.Hole AR. Modelling heterogeneity in patients' preferences for the attributes of a general
practitioner appointment. Journal of Health Economics 2008;27:1078-1094.
25.Guo N, Marra CA, Fitzgerald JM, et al. Patient Preference for Latent Tuberculosis Infection
Preventive Treatment: A Discrete Choice Experiment. Value in Health 2011;14:937-943.
26.Sadique MZ, Nancy D, Edmunds WJ, et al. The Effect of Perceived Risks on the Demand for
Vaccination: Results from a Discrete Choice Experiment. Plos One 2013; 8:e54149.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3666226